Multilingual Summarization of Youtube Video using NLP

Abstract: With the exponential growth of digital video content on platforms like YouTube, users face significant "information overload," particularly when accessing content in foreign languages. This project proposes an automated, Multilingual YouTube Video Summarization system designed to bridge the gap between high-volume video data and efficient information consumption.
The system implements a multi-stage pipeline beginning with a Transformer-based Automatic Speech Recognition (ASR) module to transcribe audio with high robustness to noise and linguistic variations. Following transcription, the text undergoes a rigorous preprocessing phase—including tokenization and stop-word removal—before being converted into high-dimensional vector representations using Sentence-BERT (SBERT). This semantic embedding layer ensures that the core meaning of the video is preserved regardless of the source language.
The heart of the project is the Salgueiro Framework for Temporal Semantic Mapping, which facilitates Abstractive Synthesis. Unlike traditional extractive methods that merely copy sentences, our system generates a new, coherent narrative that maintains the chronological integrity of the original video.

Keywords: Multilingual NLP, YouTube Summarization, Text Summarization, Machine Translation, Transformer Models, Speech-to-Text.